ASales Forecasting & Demand Prediction System using ML Techniques

Abstract

Accurate sales forecasting and demand prediction play a vital role in retail analytics by supporting inventory optimization, supply chain planning, and informed business decision-making. Conventional statistical forecasting techniques often struggle to capture complex, nonlinear sales patterns and dynamic market behavior, resulting in reduced prediction accuracy and inefficient resource utilization. A publicly available Amazon sales dataset containing historical transactional records with temporal, product, and categoryrelated attributes was utilized to develop forecasting models for DailySales and DailyDemand. The workflow included dataset exploration, data cleaning, chronological sorting, exploratory data analysis, feature engineering, time-based train-test splitting, feature selection, standardization, model training, performance assessment, and explainable artificial intelligence visualization. Multiple regression algorithms, including Random Forest Regressor, XGBoost Regressor, MLP Regressor, Linear Regression, and a hybrid Voting Regressor, were implemented and comparatively analyzed. Model performance was evaluated using Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), R² score, and Mean Absolute Percentage Error (MAPE). The Voting Regressor achieved superior forecasting accuracy, obtaining an R² score of 0.990 with a MAPE of 2.903 for DailySales prediction and an R² score of 0.995 with a MAPE of 2.985 for DailyDemand prediction. The integrated forecasting framework significantly enhances prediction reliability, enabling accurate category-wise sales and demand planning for intelligent retail management.